Issue with 12V to 230V full bridge 1.5kW converter

[edgaras006]

Hello everyone,

I have been battling with 12V to 230V converter intermittently over a month now. It should take from 10.3V to 14.4V and convert it to 230VDC up to 6.5A. I know that layout is critical in this design. PCB is 6 layers 2Oz. Design uses local gnd and power planes as a reference plane along with lots of capacitors.

My issue is - it does not start up. I get 40V on the output, U14 conduct almost nothing, despite having over 20V (because of unloaded transformer which generates that isolated voltage) between cathode and anode, but just 0.4V on reference pin. What have I missed?

P.S. two components with || before their value were not in original schematic. Before I added them it didn't even attempted to start.

Thank you.

Edited:
Transformer is 1:31:31, primary inductance is 4.5uH, switching frequency 110KHz. Primary is rated for 300A with DC_R of 0.35mOhm. Leakge inductance about 10nH. I rated it between 1.5nH and 25nH, custom transformer was about 10nH.

[T3sl4co1l]

R47 needs a pulldown near it.

Is supply switching really the best way to "enable" U13?  Surely you could pull up CS or something like that?

5Hx and 12Hx are absent, or at least I can't see where they are. Schematic isn't searchable, it's flattened graphics, well mostly, some text labels survive but not the important ones like labels and part numbers.

Also weird that none of the lines line up.  I assume everything is snapped to grid and connected, and that's some weird output generation glitch.  At least in my PDF reader.

Just two pins to that poor transformer? I'm very interested in how you intend to get up to 180A through that poor thing without it fusing open...  On the upside, leakage doesn't matter much, in the H-bridge configuration.

Supply, you probably want some electrolytics in there for damping, and/or TVS for transient limiting.  I'm guessing load dump isn't part of the requirement (but beware if that battery is charged by an alternator) but there's still inrush.  At least, I'm assuming since no precharge circuit is shown.

U11 should have a bias resistor in parallel with the diode, to supply D5+U14 leakage/bias currents.

C12 got turnt.

Why 1200V diodes for a 230V output? Is there a ton of transient voltage there? Why not FWB rectifier? Would snubbers help?

Tim

[schmitt trigger]

Similar complaints here.
I am attempting to follow your schematic on a cellphone’s screen, and I can’t find U14, which apparently provides some important bias voltages.

[T3sl4co1l]

TL431, right of the controller.

Not sure if you have text search on your cellphone, but yeah, hard to find things.

Schematic readability can be improved by separating and sectioning things better; all the stuff around the controller just kind of blobs together.  Doesn't have to be rigid hierarchical blocks or dividing lines, but just putting in a little space, is the direct schematic equivalent of gently spaced paragraphs.

(I think I found everything I needed to see?  I might not push back on the schematic layout in a formal design review, but the fact that it would take longer than usual, and be less accurate, would be worth noting on the review.)

Tim

[edgaras006]

Yea, schematic is messed up. I noticed that only when I zoomed in. Something terribly wrong happened to DIPTrace as it used to work perfectly in the past.

Anyway, I added 4.7K bias resistor to U11 diode and got 53V on the output instead of 41V before. Good start, I was measuring what is happening on U14 (TL431) when board started buzzing and current hit limiter (3A). OK, I switched to 25A lab power supply. For a several seconds it stays at 5.5A, before it hit current limiter too. Q20 and Q4 warm up. I have massive planes so even if there were a puch through it would take a while to warm up reasonably. It takes about 15-20 minutes to heat board up to soldering temperatures with two heatguns...

To my surprise input voltage from power supply jumps by 2V (resonating power line?)

Anyway, something is happening.

T3sl4co1l expressed concerns about contacts of transformer primary, well it is 4 16mm^2 multi strand copper windings in parallel, so that should do it.

Thanks for all of your comments, up to this point it really helped. I had never done design this powerful before. I had few hundred watts step down from mains or step up designs, but all below 500W. Some of them were zero voltage switching full bridges, but this one is a beast for me which I need to tame.

Apparently I have a puch through even if gate channel resistance looks OK'ish, but channel is shorted Q20.

[xavier60]

What does C42 do? looks like it could cause instability.
Minimum load current for the TL431 before it's allowed to control the loop is 1mA, usually provided by a 1K resistor across the LED.

[xavier60]

Is C42 meant to be the "fast lane" path?
Page 12,  https://www.onsemi.com/pub/Collateral/TND381-D.PDF

[moffy]

U11 should have a bias resistor in parallel with the diode, to supply D5+U14 leakage/bias currents.

Tim

As T3sl4co1 mentioned about a parallel resistor and you used a 4.7k, that might be a little large as the TL431 can have up to 400uA leakage before conducting and the input diode to the opto coupler has around 1V voltage drop for 0.1ma. You might want to drop the value to 2k or less.

[edgaras006]

xavier60>>> Yes, exactly that.

[xavier60]

I think C42 would be better connected to the LED's anode and the TL431 made slower. I don't exactly know how to optimize compensation except by experimenting.

[edgaras006]

I will show my shortsightedness, but somebody might benefit from that. My thinking was, OK if TL431 starts up through LED, proven that output resistor is low enough it will be OK if that happens within soft start period. It turns out I can get away with that with low power converters, but not this one. For high current converters TL431 needs to give reliable feedback all the time. Start up through LED is not an option. So whoever is reading this, always add 1K in parallel to optocoupler LED, so TL431 can start up properly and then start giving feedback.

After lowering resistor parallel to LED to 1K it no longer draws tens of amps on startup, rather then 0.6A. Another issue, gate drivers supply. I have small push pull transformer for that capable of sourcing 150mA at 12V. It looks like it is too weak for 30A gate driver drawing 120mA average current and 82uF 16V electrolytic sitting on primary side of push pull transformer heats up like crazy from pulse loading. As a backup option I will install additional isolated 2W power supplies for upper transistors bias voltages (I intentionally placed them on PCB in case of situation like that). We will see. I assumed that having 6x4.7uF in 1206 will provide enough pulse current for them. I appears I was wrong. I continue to debug it and hopefully find the problem soon.

[T3sl4co1l]

Yeah, C42 is in the wrong place, it's adding a massive zero to the transfer function.  TL431 starts up within microseconds, it's ready and stable within a switching cycle, the controller doesn't know anything different.  Just ensure +10Va_HV is available before enabling -- a UVLO on that, and the high sides and whatnot, might be handy to have.

Specific variant BTW is a low current type,
https://www.ti.com/lit/ds/symlink/atl431.pdf
so not much bias should be needed.  You still need proper compensation (once it gets close enough to nominal voltage, anyway), which C45 will not do, but an R+C can be patched in there just fine.


...So, you aren't taking any waveforms here?  Several of these items would be easily addressed by just probing at things.  (Not sure if/how/why a reminder is necessary when working on a manifestly time-domain system like an SMPS...)

If you don't have a scope, I'm not sure what hope you had of working with this.  That's practically, if it works it works and you're insanely lucky, and if not you're flailing in the dark forever.

Mind, even with a scope, managing currents and temperatures will be a nightmare -- those multiple strands in the primary certainly will not share evenly, as an example -- but that's a scaling problem, not the immediate basic-operation problem.

Tim

[edgaras006]

T3sl4co1l>>>I have RTB2004 and it helps tremendously, definitely along with IT8615 and IT8512C electronic loads and a buch of other gear, I just was not sure what a heck is going on with either TL431 or controller, because there was steady +-200mV ripple at 55KHz on the output and TL431 was not behaving no matter what I do. Transformer primary saw +-4V (almost sine). Now I have some action and waveform which allow seeing what is going on. That TL431 was real issue for me.  When I resolve this I will post an update and if I bump into something else there will be some waveforms to share.

[schmitt trigger]

Hopefully you have a diff probe such that waveforms of upper and lower Mosfets can be viewed simultaneously. Just to check for correct dead times .

[edgaras006]

Hopefully you have a diff probe such that waveforms of upper and lower Mosfets can be viewed simultaneously. Just to check for cross conduction.

Three of them. I have some experience with isolated power supplies, but not much. Majority of my work is analog front ends, high speed digital, EMI, and non isolated converters powering variuos sub components of mixed signal circuits. I recently started working with off line converters and UPS'es.

[edgaras006]

I noticed that local 12V 10uF cap supplying power to gate drivers is getting hot. Measured gate drive wave forms and low side looked bad, signal on gates never got above 5V while they should reach 12V. I added massive 4700uF cap (98mOhm) and wave forms become OK. However my bench power supply still complains. I attempted to start up supply with 13.8V and 5A current limit. Not good. I increased current limit to 28A, still the same. I think there is some issue. When calculating transformer its magnetization current was about 35A, but I assumed that local caps will supply charge required to magnetize core and boost voltage on secondary if secondary is not loaded. I have 100A 12V power supply, but it does not have current limit, so I am hesitant to use it before I have stable output voltage without any load.

What could be my next steps?

Waveform in the oscilloscope captures shows voltage on input capacitor 4700uF 16V 98mOhm. Right next to gate drivers there are 24x4.7uF 16V 1206 X7R caps and on the entry to the board there are 25x4.7uF 16V 1206 X7R before 100uOhm current sense resistor.

[xavier60]

Looks like an overload causing the 12V to be pulled down to the 9V UVLO threshold. More shorted Gates?
The drives should be checked before main 12V is supplied to the MOSFET bridge.

[Vovk_Z]

. I added massive 4700uF cap (98mOhm) and wave forms become OK.

Don't you forget about existing polymer caps? They have really low ESR and work pretty well with large current spikes. So you don't need large rated capacitance and still have very low ESR.

[edgaras006]

Double face palm  I fallowed reference schematic on the datasheet which was wrong. Both low and high sides are switching together.

Overcurrent was caused by soldering error as well as previous 40V output at not load (one floating gate)

[T3sl4co1l]

Huh, are your probes poor quality? Or is there a high frequency compensation adjustment that's been neglected?  Strange seeing the "droop" artifact at this time scale, and I don't see why it would show up in the circuit otherwise.

You might want to zoom in and investigate the transient (edges and peaks).  Or determine if that's also a probe artifact as well, or what.

Tim

[edgaras006]

Huh, are your probes poor quality? Or is there a high frequency compensation adjustment that's been neglected?  Strange seeing the "droop" artifact at this time scale, and I don't see why it would show up in the circuit otherwise.

You might want to zoom in and investigate the transient (edges and peaks).  Or determine if that's also a probe artifact as well, or what.

Tim

Q20 Lower right leg of H bridge - non differential RT-ZP03 probe which was a little overcompensated.
Q14 Lower left leg of H bridge - differential 1.4kV probe Cal Test 2593-1 is slightly undercompensated for high frequency and there is no compensation pot.
Other two channels Micsig DP10007 1kV differential probes, also no way to compensate it.

How to do compensate diff probe? I do not want to open it, it might be ultrasonically welded.

[T3sl4co1l]

Aha, the HF trimmed kind, that would do it.

CT2593-1 says 50...uhhhh, kΩ source impedance and load??  Surely that's a typo and they mean 50Ω?  Well, it's obsolete now I guess, so unless you have a printed manual somewhere, that might be about all there is to know about it anymore, shrug.  Does it have rated gain (is it off by 2x)?  Does it read better with 50Ω termination enabled?  At 25MHz BW, it's never going to read as crisp as the passive probe, of course.

Refs:
https://scdn.rohde-schwarz.com/ur/pws/dl_downloads/dl_common_library/dl_manuals/gb_1/r/rt_zp/RT-ZP03_UserManual_de_en_04.pdf
https://mm.digikey.com/Volume0/opasdata/d220001/medias/docus/4846/DS_1768_Catalog.pdf p.8

Tim

[edgaras006]

I swapped A and B places of U13 and still nothing good is coming from that.

SCR68 bow give correct MOSFET switch on and off
SCR69 zoomed out start up showing output
SCR71 zoomed in start up showing output

I also tried to drive transformer using signal generator (50Ohm output).
SCR72 generator tried outputting sine.
SCR74 generator was set to output square wave.

Well transformer is working. Can somebody see what have I missed?

Edited: CT2593-1 with 50Ohm termination didn't help at all, just amplitude collapsed. It looks like its output is about 75Ohms.
Edited 2: I must have been blind. From current configuration A and C have to be swapped.

[edgaras006]

Made it. After carefully checking all connections, I had to swap AB lines as I had done and had soldering error. Cold joint. It is not that easy to heat 12Oz copper (6x2Oz) and solder DFN on top.

[edgaras006]

I have an update on the issues I had. The datasheet contains three nasty mistakes I had found so far.

The First issue and the reason why I created this thread is obvious: mislabeled pins. Right after I measured them it is clear - A and B needs to be swapped places. Yes, there is a line in the datasheet describing them correctly, but not on a section which describes pins, whats more reference schematic contains the same mistake.

The second issue: Missing information about a resistor necessary to prevent core saturation at light load. This information is published in SLUA275, yet there is not a word about it in the main datasheet.

The third issue: incorrect or even false information about compensation scheme in peak current mode in which I want controller to operate.

Lastly (icing on the cake): I had found a texas instruments forum thread discussing issues with a client which offered to write organized paper describing all the error he had found in UCC2895 datasheet and TI just didn't respond to that. I hope my struggles with help somebody someday.

Anyway, I will attempt to compensate it and try to stick with companies which put more attention to documents (Intersil, Maxim (now AD), Linear Tech (now AD), NXP). This chip was orginally designed by unitrode and its documentation is not upto Ti standards.